This work presents an integrated batch-to-batch control and within batch re-optimisation control strategy for batch processes using neural network models. To overcome the difficulties in developing detailed mechanisti...
详细信息
ISBN:
(纸本)0780386353
This work presents an integrated batch-to-batch control and within batch re-optimisation control strategy for batch processes using neural network models. To overcome the difficulties in developing detailed mechanistic models, neural network models are developed from process operation data. Due to model-plant mismatches and unknown disturbances, the optimal control policy calculated based on the neural network model may not be optimal when applied to the actual process. Utilising the repetitive nature of batch processes, neural network model based iterative learning control is used to improve the process performance from batch to batch. Batch-to-batch control can only improve the performance of the future batches. Within batch re-optimisation should be used to overcome the detrimental effect of disturbances on the current batch. The proposed technique is successfully applied to a simulated batch polymerisation process.
The perfonnance of a control chart in statistical processcontrol is often quantified in terms of the Average Run length (ARL). The ARL enables a comparison to be undertaken between various monitoring strategies. Thes...
详细信息
The perfonnance of a control chart in statistical processcontrol is often quantified in terms of the Average Run length (ARL). The ARL enables a comparison to be undertaken between various monitoring strategies. These are often detennined through Monte Carlo simulation studies. Monte Carlo simulations are time consuming and if too few runs are perfonned then the results will be inaccurate. An alternative approach is proposed based on analytical computation. The analytical results are compared with those of the Monte Carlo simulations for three case studies.
This paper proposes the application of Gaussian process regression for the empirical modelling of batch processes to provide long range predictions. Gaussian processes are flexible, non-parametric Bayesian regression ...
详细信息
This paper proposes the application of Gaussian process regression for the empirical modelling of batch processes to provide long range predictions. Gaussian processes are flexible, non-parametric Bayesian regression techniques. In the training stage, hyper-parameters that define the covariance structure of the Gaussian process can be obtained using Markov Chain Monte Carlo sampling. Model predictions can then be achieved by taking the average of the Monte Carlo samples. The proposed technique is evaluated by application to a benchmark simulation of a fed-batch bioreactor. The results show that comparable results can be achieved with other non-parametric modelling approaches such as recurrent neural networks.
In a changing operational environment, a major challenge that still exits is the assured state and parameter estimation of dynamic processes. The values or expressions of important parameters can be difficult to deter...
详细信息
In a changing operational environment, a major challenge that still exits is the assured state and parameter estimation of dynamic processes. The values or expressions of important parameters can be difficult to determine and initial errors may be present in some parameters as a result of changes in the initial operating conditions. Furthermore as a result of variations in the environmental and operational conditions or the dynamic characteristics of the process, parameters may be time-varying. In this paper, an on-line Bayesian parameter estimator is developed and evaluated on a simulation of a batch methyl methacrylate process.
Successful application of model based control depends on having good estimates for the system dynamic states and parameters. A multivariate dynamic linear model is developed for the estimation of the states from limit...
详细信息
Successful application of model based control depends on having good estimates for the system dynamic states and parameters. A multivariate dynamic linear model is developed for the estimation of the states from limited measurements in a non-linear system comprising model uncertainties. Since the noise statistics are rarely available a priori, the noise covariance matrix is treated as a tuning parameter and determined through repeated simulations. For non-linear, time varying processes, the assumption of a constant process noise covariance matrix does not realise accurate estimates. In this paper Monte Carlo simulations are used to obtain the time-varying noise covariance matrix. The methodology is demonstrated on a benchmark polymerisation process.
Multi-way statistical projection techniques have typically been applied in the development of monitoring models for single recipe or single grade production. As defined, implementation of these techniques in multi-pro...
详细信息
Multi-way statistical projection techniques have typically been applied in the development of monitoring models for single recipe or single grade production. As defined, implementation of these techniques in multi-product applications necessitates the development of a large number of process models. This issue can be overcome through the use of common sub-space models constructed by pooling the individual variance-covariance matrices. A second issue with multi-way approaches is the difficulty of interpreting multiway contribution plots. An alternative approach is the U2 statistic. In this paper an extension is proposed, the V2 statistic, based on the cumulative contribution of variables at each sample point. The methodologies are demonstrated on two industrial applications.
Partial Least Squares (PLS) is a popular method for the development of a framework for the detection and location of process deviations. A limitation of the approach is that it has generally been used to monitor one r...
详细信息
Partial Least Squares (PLS) is a popular method for the development of a framework for the detection and location of process deviations. A limitation of the approach is that it has generally been used to monitor one recipe, one product, for example, consequently applications may have been ignored because of the need for a large number of process models to monitor multi-product production. This paper introduces two extensions - multi-group and multi-group-multi-block PLS. These techniques enable a number of similar products, manufactured across different unit processes, to be monitored using a single model. The methodologies are demonstrated by application to a multi-recipe industrial manufacturing process.
Calibration-free resolution techniques provide an alternative approach to the development of a calibration model. These combine spectroscopic measurement coupled with mathematical and statistical assumptions and give ...
详细信息
Calibration-free resolution techniques provide an alternative approach to the development of a calibration model. These combine spectroscopic measurement coupled with mathematical and statistical assumptions and give spectral profiles and non-quantitative concentration profiles for the unknown mixture. In this paper, a number of calibration free techniques including VARIMAX. ITTFA, EFA, FSWEFA, SIMPLISMA are described and applied to a synthetic spectral data set and the results are compared with the complementary technique of Independent Component Analysis (ICA) in particular FastICA and JADE. The results were comparable in all cases with ICA separating the signal from the constituent components successfully.
This paper presents two soft-sensing models for predicting the product yields profile and the cracking degree of an ethylene pyrolysis furnace. The model based on single neural network with only one hidden layer train...
详细信息
ISBN:
(纸本)0780386531
This paper presents two soft-sensing models for predicting the product yields profile and the cracking degree of an ethylene pyrolysis furnace. The model based on single neural network with only one hidden layer trained by Levenberg-Marquardt algorithm with regularisation was first developed. It was found that the single neural network lack generalisation capability in that they can give undesirable performance when applied to unseen data. To improve the generalisation capability of the soft-sensing model, multi-model soft-sensors based on bootstrap aggregated neural networks with sequential training are used. In the sequential training of bootstrap aggregated networks, the first network is trained to minimise its prediction error whereas the rest of the networks are trained not only to minimise their prediction errors but also minimise the correlation among the trained networks. The overall output is obtained by combining all the individual networks. Application results show that the multi-model soft-sensors possess good generalisation capability in that they give good performance when applied to unseen data.
A challenge facing the pharmaceutical and chemical industries is how to understand and identify differences in process behaviour where a product is manufactured at two different sites. Three approaches based on multi-...
详细信息
A challenge facing the pharmaceutical and chemical industries is how to understand and identify differences in process behaviour where a product is manufactured at two different sites. Three approaches based on multi-group principal component analysis are investigated and benchmarked against single site models. The multi-group approach is shown to remove differences between sites such as operational scale thereby enabling the analysis to focus on identifying differences in variation between the two sites that are not a consequence of process configurations. From the analysis it is observed that the multi-group approach can assist in the understanding of manufacturing performance.
暂无评论